Adam H. Monahan

Physically based observational constraint methods can effectively reduce uncertainty in global warming projections but have not been widely applied at regional scales. We first develop and apply multivariate linear regression models for constraining projections of surface air temperature averaged over subcontinental regions in the extratropical Northern Hemisphere, based on a set of potential constraints including climatological metrics derived from tropical and subtropical low-level cloud and global average past warming trend, as well as a set of regional climate metrics previously used in the literature. We evaluate the performance of the multivariate linear regression models based on cross-validated tests using output from phases 5 and 6 of the Coupled Model Intercomparison Projects (CMIP). We find that linear regression models using global-scale low-cloud metrics alone perform more robustly than linear regression models using the past global mean warming trend or regional climate metrics as constraints. These results, while favoring global constraints over the set of regional constraints considered, do not preclude the existence of even better regional constraints for particular regions. Through model-based cross-validation, the projections constrained using low-level cloud metrics exhibit more accurate best estimate projections, narrower uncertainty ranges, and more reliable uncertainty estimates in most Northern Hemisphere regions when compared with unconstrained projections. Application of the approach to climate projections based on both Shared Socioeconomic Pathway (SSP) 1-2.6 and SSP5-8.5 using observed low-cloud metrics results in considerably narrower 5%–95% uncertainty ranges of twenty-first-century warming over subcontinental Northern Hemisphere land regions compared to unconstrained projections.